An Empirical Evaluation of Explainable AI for Vulnerability Detection and Localization